aws-finops-bridge

MCP bridge that lets IDEs (VS Code / Cursor) talk to the AWS FinOps Assistant backend.

What it does

- Runs as an MCP server over stdio (what IDEs expect).
- Forwards tool discovery/calls to your remote backend over HTTP (via the backend /messages endpoint).

Environment variables

Required:
- MCP_SERVER_URL — your backend base URL (can be the Lambda Function URL). If it doesn’t end with /sse, the bridge will append /sse.
- APP_API_KEY — the Application API key.
- AWS_ACCESS_KEY_ID
- AWS_SECRET_ACCESS_KEY

Optional:
- AWS_SESSION_TOKEN
- AWS_REGION (defaults to us-east-1)

Run locally

``bash
MCP_SERVER_URL="https:///sse" \
APP_API_KEY="..." \
AWS_ACCESS_KEY_ID="..." \
AWS_SECRET_ACCESS_KEY="..." \
AWS_REGION="us-east-1" \
node ./index.js
`

Publish to npm (public)

1) Make sure the package name is available:

`bash
npm view aws-finops-bridge
`

If it’s taken, use a scoped name like @your-scope/aws-finops-bridge.

2) Login to npm:

`bash
npm login
`

3) From this folder, publish:

`bash
npm publish --access public
`

4) Test without installing:

`bash
npx -y aws-finops-bridge
`

IDE config (MCP)

After publishing, a typical MCP config uses npx to run the bridge:

`jsonc
{
"servers": {
"aws-finops": {
"command": "npx",
"args": ["-y", "aws-finops-bridge"],
"env": {
"MCP_SERVER_URL": "https:///sse",
"APP_API_KEY": "${input:APP_API_KEY}",
"AWS_ACCESS_KEY_ID": "${input:AWS_ACCESS_KEY_ID}",
"AWS_SECRET_ACCESS_KEY": "${input:AWS_SECRET_ACCESS_KEY}"
}
}
}
}
``